fscrypt: document gfp_flags for bounce page allocation

Document that fscrypt_encrypt_pagecache_blocks() allocates the bounce
page from a mempool, and document what this means for the @gfp_flags
argument.

Link: https://lore.kernel.org/r/20191231181026.47400-1-ebiggers@kernel.org
Reviewed-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Eric Biggers <ebiggers@google.com>

diff --git a/fs/crypto/crypto.c b/fs/crypto/crypto.c
index fcc6ca792ba2c75decf0c1e1f1af9a60dc0085f3..1ecaac7ee3cb8e0ad958fd9b935eb4d4e0169f23 100644 (file)
--- a/fs/crypto/crypto.c
+++ b/fs/crypto/crypto.c
@@ -138,7 +138,7 @@ int fscrypt_crypt_block(const struct inode *inode, fscrypt_direction_t rw,
  *             multiple of the filesystem's block size.
  * @offs:      Byte offset within @page of the first block to encrypt.  Must be
  *             a multiple of the filesystem's block size.
- * @gfp_flags: Memory allocation flags
+ * @gfp_flags: Memory allocation flags.  See details below.
  *
  * A new bounce page is allocated, and the specified block(s) are encrypted into
  * it.  In the bounce page, the ciphertext block(s) will be located at the same
@@ -148,6 +148,11 @@ int fscrypt_crypt_block(const struct inode *inode, fscrypt_direction_t rw,
  *
  * This is for use by the filesystem's ->writepages() method.
  *
+ * The bounce page allocation is mempool-backed, so it will always succeed when
+ * @gfp_flags includes __GFP_DIRECT_RECLAIM, e.g. when it's GFP_NOFS.  However,
+ * only the first page of each bio can be allocated this way.  To prevent
+ * deadlocks, for any additional pages a mask like GFP_NOWAIT must be used.
+ *
  * Return: the new encrypted bounce page on success; an ERR_PTR() on failure
  */
 struct page *fscrypt_encrypt_pagecache_blocks(struct page *page,